Automations | Yarken Help

Automations run configured AI workflows on a schedule, on an app event, from a webhook, or manually. A workflow can read data, evaluate conditions, request approval, send notifications, refresh cubes, and create findings for review.

Use them when a repeatable process needs to run consistently without rebuilding the same sequence in Ask Yarken each time.

When to use Automations

Use case	Typical trigger	Workflow
Investigate a spend anomaly	On app event: an insight fires	Read insight, check type and severity, explain the finding, then raise an action
Refresh reporting data	Schedule	Select cubes, refresh pending partitions, and review run output
Generate periodic review	Schedule or Manual	Run saved report, summarize results, notify recipients
Process an authorized external event	Webhook	Validate trigger configuration, read payload, run selected steps

Create an Automation

Open Admin > AI Studio > Orchestrate > Automations.
Create an Automation and give it a name describing the job, such as Daily cloud anomaly review.
Select the Trigger on the canvas. Choose a trigger type and complete its configuration.
Set Data access based on which records the workflow must read.
Select + Add step to build the workflow. Select a node to edit its fields in the right panel.
Select Save. Use Test run, review Run History, and activate the Automation after checking the outcome.

Choose a trigger

Trigger	Use it when	Key setting
Schedule	Work should repeat at set times	Recurrence and time
On app event	An event in Yarken should start a workflow	Event type and applicable insight rules
Webhook	An external caller should trigger a workflow	Token and allowed IP addresses, as configured
Manual	Admin runs workflow on demand	Manual execution

For On app event > An insight fires, the Insight rules selector can narrow the events that start the flow. An empty selection matches every insight for that event.

For Schedule, set the recurrence, day, time, and timezone. For Manual, the workflow runs only when an administrator starts it.

Data access: My access only uses the workflow owner's authorization. Organization-wide runs with broader configured service-account access. Choose organization-wide only when the workflow genuinely requires broader scope and that access is approved.

Important: Data access affects every read and write that follows. Test with the minimum scope needed before activation.

For external triggers, see Configure a webhook-triggered automation.

Build the workflow

Most Automations combine a small set of reusable building blocks: variables, instructions, structured extraction, conditions, loops, and summaries.

Use variables between steps

Variables pass values from one step to another or keep date-based filters current without editing the Automation each period.

Select variable, or type { in supported fields, to choose an available workflow value.
Use Store as variable on supported steps when later steps need that output.
Common values include {{today}}, {{current_year}}, {{current_month}}, {{current_financial_year}}, and {{last_output}}.
Make sure the variable resolves to the field type the next step expects. Calendar and financial-year values are different periods.

Use Instruction, Extract values, and Summarize

Use Instruction when the agent needs to analyse or transform scoped information.
Add Extract values when a later step needs structured fields such as an amount, reason, identifier, or table.
Set a clear variable name and type for each extracted value.
Use Summarize after the values are validated. Choose AI summary for generated wording or Template for fixed text with inserted variables.
Use Set variable when you need a fixed value or a direct copy of another variable.

Do not connect an unverified extracted value directly to a write, notification, or action step. Test missing values and type conversion first.

Use conditions and loops

Use Condition to choose a path and For each to repeat steps for each row in a table.

Condition: use Compare values for explicit logic or Let AI decide when the branch depends on an AI-evaluated condition.
The first matching branch runs. Add Otherwise when unmatched input needs a defined outcome.
For each: select a compatible table output, assign a short item name such as item, then reference current-row fields inside the loop.
Test loops with a small dataset before allowing them to create or update multiple records.

Choose the steps you need

Generic builder concepts are covered on this page. The linked child articles below are kept for steps with distinct configuration, permissions, write behavior, or troubleshooting requirements.

Step	What it does	Help article
Instruction	Gives agent a task to perform	See Build the workflow above
Extract values / Set variable	Captures results for later steps	See Build the workflow above
Summarize	Produces AI-generated or template output	See Build the workflow above
Condition / For each	Branches on a test or processes each item	See Build the workflow above
Wait for approval	Pauses flow for human review	Require approval
Notify	Sends configured message to recipients	Send notifications
Read / Update insights	Reads an insight or changes supported insight data	Read and update Insights
Run report	Executes selected saved report	Run a saved report
Refresh cubes	Queues selected cube refresh work	Refresh cubes
Read / Write assets	Reads or updates asset records	Read and write assets
Business case steps	Reads or updates business-case information	Read and update business cases
Recommendation steps	Reads, writes, or changes recommendation status	Read and update recommendations
Write companion metric	Creates or updates an associated metric	Write companion metrics
Raise an action	Creates an assigned finding from a configured Action Type	Raise an action

From finding to follow-up: two Automations

Start in AI Settings. Create Action Type at Admin > Settings > AI > Action Types before adding a Raise an action step. Action Type defines fields, card presentation, recipient, and repeat-finding identity. See Action Types.

Producing Automation: Use On app event > An insight fires, a schedule, or other supported trigger. Read data, evaluate condition, extract values, then add Raise an action and map Action Type contract fields. This creates assigned action card.
Follow-up Automation: Create another Automation. Set trigger to On app event > An action is accepted. Under Which action types?, select Action Types it should handle. Add steps to investigate, prepare reports, notify users, or raise another action. Save, test, and activate.
Recipient action: In Inbox, open card and select Act on this. This triggers matching follow-up Automation. Check its Run History to confirm execution.

Acceptance and approval serve different purposes. Act on this starts configured follow-up work. For a separate authorization decision, add Wait for approval before consequential changes. Action Types do not create or connect Automations automatically.

Detailed instructions: Raise an action from an Automation and Review findings with Take Action.

Example: Turn an insight into an assigned action

Choose On app event > An insight fires and select relevant insight rules.
Add Read insights to load the triggering record.
Add Condition, such as insight.type == anomaly, to process only the intended class of event.
Add Instruction to draft a concise finding based on the insight's data. Then use Extract values for the card headline and evidence.
Add Raise an action, select an existing Action Type, and map its required fields, headline, and evidence to the values extracted earlier.
Test the workflow. Confirm it creates the intended action and check the recipient and impact value.

Action Types are configured under Admin > Settings > AI > Action Types. Each type has its own payload field names. Under its Lifecycle settings, select stable fields to identify repeat findings. Avoid selecting changing measures when repeated runs should update the existing action.

See Action Types and Review findings with Take Action.

Example: Refresh reporting data weekly

Set a Schedule trigger.
Add Refresh cubes, then select one or more cubes.
Choose Only pending partitions when the goal is to refresh changed data rather than rebuild every selected cube in full.
Save and test. Check Automation Run History and Admin > Cube refresh to verify results.

Refresh cubes queues work; the Automation continues without waiting for the refresh to finish. A following report may still show prior values if it runs before the refresh completes.

Test and investigate runs

Use Test run before activating a new or changed Automation.
Open Run History and inspect total runs, success rate, duration, and each run's status.
For webhook Automations, open Blocked Requests when a caller did not start a run. Check received time, source IP, and reason.
Correct configuration, then run the test again.

Blocked webhook requests can report IP address isn't on the allowed list or Wrong or missing token. These requests do not start a workflow run. See Investigate automation runs and blocked webhooks.

Before activation

Check that each step receives the fields and types it expects.
Use Wait for approval before sensitive writes or communications that require review.
Confirm recipients, action ownership, and repeat-finding fields.
Verify report authorization and the minimum data-access scope needed.
Check the latest test results and any queued work before enabling recurring execution.

Related content